Tool Condition Prediction in High-Speed Milling Using Deep Q-Networks (DQN) and Sensor Fusion Data

Highly effective and reliable high-speed milling depends on the timely tool condition monitoring (TCM) check. In this paper, we propose a novel approach for tool condition prediction based on DQN with sensor fusion data. Both contact and non-contact sensors such as, vibrational, force, and acoustic emission sensors are utilized to gather data in real time during milling. The information collected by these sensors are integrated into a number of important features to be used in DQN model to learn the prediction of tool wear and possible failure. The DQN model is built to learn from the response of the milling environment characterized by tool condition in terms of wear and the remaining useful life (RUL). The results derived from experiments tend to show that the utilised DQNbased model has better quantitative values for accuracy, precision, recall, and F1-score than classical machine learning models like artificial neural networks (ANN), support vector regression (SVR), decision trees and etc. The integration of the sensor data enhances the overall prediction reliability, a factor which is highly achieved from the fusion of the data received from the vibration, force and acoustic emission sensors. The overall system proposed has significant application for real time implementation and demonstrates relatively short prediction time and is less sensitive to changes in the machining conditions. This proposed methodology is a feasible approach for improving tool life and reducing time needed for high speed milling applications.